Development progress of N-PET, a high-sensitivity and high-resolution TOF-DOI PET prototype dedicated for human brain imaging

To achieve high sensitivity and high spatial resolution for human brain PET imaging, the N-PET project is developing a range of methodological advancements from the innovative multi-resolution TOF-DOI detector to multiple machine learning methods that can enhance DOI discrimination and improve DOI resolution. We aim to develop an N-PET prototype system that offers a ${\lt}1.5 \mathrm{~mm}$ spatial resolution, at least 4 levels of DOI discrimination and ${\lt}300 \mathrm{ps}$ TOF resolution. The major innovations include a multi-resolution TOF-DOI detector with two-layer LYSO crystal arrays and lightguide that can achieve high spatial and TOF resolution as well as DOI decoding capacity, an efficient and practical DOI discrimination method at the system level using semi-supervised learning to reduce labeling costs while maintaining reliability, and a CNN method trained by detector side-irradiation data acquired at different depths that can further improve DOI resolution by 48%. Currently, preliminary experimental results show that the detector is able to achieve a TOF resolution of 260 ps and a DOI resolution of better than 4.2 mm at bottom layer. The impact of TOF and DOI information on image quality is evaluated using GATE Monte Carlo simulation. A specialized image quality phantom suitable for evaluating brain-dedicated PET systems has been designed, and image reconstruction and normalization algorithms has been developed. The simulation results demonstrate that the imaging quality of the multi-resolution detector-based system is enhanced by using its TOF and DOI information. The development of a single-ring prototype is in progress, with more detectors and electronics are being delivered. We plan to present the prototype’s initial performance at the conference, including spatial resolution, CTR, and results from phantom and animal imaging.